6 Results
6.1 Differing welfare rationales
Before commencing with the results of the first part of the analysis, I will first report on the results of the preliminary factor analysis conducted beforehand (see Appendix 1.1). For both the Varimax and the oblique rotation, the Kaiser-Meyer-Olkin measure (KMO) confirmed the sampling adequacy of the factor analysis with a value of KMO = 0.89 respectively.17 Regarding the individual items, the KMO was greater than 0.782 and therefore above the acceptable limit of 0.5 (cf. Field 2017: 600). The three factors identified through the analysis (together explaining 56.51 % of variance) could not confirm the previous theoretical generation of factors in its entirety. It verified the construction of the factor ‘deservingness rationale’ and furthermore corroborated the elimination of items reflecting racist and ethnocentric perceptions of immigrants, which were too far removed from sociotropic cultural concerns. Apart from this, the factor analysis did not separate the items in a more concise manner reflecting the theory-based factors above. The items representing the self-interest, the economic and the cultural sociotropic rationales loaded high on the first factor only. A differentiation according to the theoretical hypotheses was thus not confirmed by the analysis. This leads to two preliminary conclusions: First, different anti-immigrant perceptions and welfare chauvinist attitudes are highly related with each other. Separating economic and cultural concerns, for example, is therefore difficult due to the oftentimes diffuse and contingent nature of these perceptions. Secondly, the items offered by the ESS-survey may not be ideally suited for analysing welfare chauvinist rationales. Consequently, between-country differences and effect sizes within the analysis are likely to be affected negatively. Any interpretation must account for this issue. Despite these difficulties with the survey-design and the underlying data set, it
may nonetheless be possible to observe a general trend within the data which warrants further study into this line of questioning. Therefore, I proceeded with the theoretical factor composition presented above for the subsequent analysis. Before commencing, though, I started by subjecting all variables to a z-Transformation to ensure comparability and the normal distribution.18
The MANOVA examining the rationales underpinning welfare chauvinist attitudes generated mixed results, due mostly to problems with the available data (see Appendix 1.2). The analysis confirmed a significant effect of welfare regime types on the different patterns of welfare chauvinist attitudes using Pillai’s trace, with V = 0,16, F(16, 48040) = 124,01 and p = 0,000 (see Table 4).19
Table 4 – Results for the MANOVA multivariate testsa
Effect Wert F Hypothesis df Error df Sig.
Intercept Pillai’s Trace 0,004 10,753b 4,000 12007,000 0,000
Wilks’ Lambda 0,996 10,753b 4,000 12007,000 0,000
Hotelling’s Trace 0,004 10,753b 4,000 12007,000 0,000
Roy’s Largest Root 0,004 10,753b 4,000 12007,000 0,000
Welfare_regime_type Pillai’s Trace 0,159 124,008 16,000 48040,000 0,000
Wilks’ Lambda 0,846 128,643 16,000 36682,629 0,000
Hotelling’s Trace 0,176 131,757 16,000 48022,000 0,000
Roy’s Largest Root 0,134 403,664c 4,000 12010,000 0,000
a. Design: Intercept + Welfare_regime_type b. Exact Statistic
c. The statistic is an upper bound on F that yields a lower bound on the significance level.
d. Computed using Alpha = ,01
Nonetheless, this result must be regarded with caution due to problems with foregoing assumption tests. Both the Levene-test20 of equality of error variances and Box’s M21 testing the equality of covariance matrices were significant at the 0,01 level (having a p-value of p=0,000 respectively, cf. Tables 5 & 6). Thus, there was no formal confirmation which let us reject the null hypothesis. Neither the elimination of multivariate outliers using Mahalanobis Distance22, nor the adjustment of group sizes through a random sampling of cases could eliminate these reservations. As both tests are highly sensitive to large sample sizes it is possible for me to disregard them both (Field 2017: 551f.). According to Tabachnick & Fidell (2013: 84f.; 253f.), when sample sizes are large and group sizes are equally distributed the robustness of multivariate significance tests like Pillai’s trace and Hotelling’s T² can be assumed. Despite these caveats I was therefore able to proceed with
18 The z-Transformation is a procedure which transforms arbitrary data int z-values, characterised by a mean of 0 and a standard deviation of 1 (Bortz and Schuster 2010: 586).
19 Pillai’s trace is on of the standard test statistics used for MANOVA, next to Hotelling’s T², Roy’s largest root and Wilks’ Lambda.
It measures the contribution of effect to the model with values (V) ranging from 0 to 1 with larger values signifying higher effect contributions. Meanwhile, the F-value (F) is the test value and determines if the variables are jointly significant. The p-value (p) decides whether or not the null-hypothesis is significant (to this see Bortz and Schuster 2010: 483; see also Statistics How To 2019).
20 The Levene-test checks the homogeneity of variances (Field 2017: 193).
21 Box’s M tests the assumption of equality of covariance matrices (Field 2017: 551).
22 The Mahalanobis-Distance is a statistical method for identifying multivariate outliers by assessing the individual distance of each case from all other cases (Tabachnick and Fidell 2013: 75).
the analysis due to the high sample size (N=12.015) and the equal group sizes (N=2.403) in the analysis. The individual effect of the respective welfare regimes suggested by the univariate test statistics (significant at p
= 0,000, see Appendix 1.2: 36) seemed to confirm this assessment.
Table 5 – Box’s Test of Equality of Covariance Matrices
Box’s M 967,755
F 24,177
df1 40
df2 318176691,177
Sig. 0,000
Tests the null hypothesis that the observed covariance matrices of the dependent variables are equal across groups.
a. Design: Intercept + Welfare_regime_type
Table 6 - Levene's Test of Error Variancesa
F df1 df2 Sig.
z-Value: Self-Interest Rationale Based on mean 106,759 4 12010 0,000
Based on median 90,227 4 12010 0,000
Based on median with adjusted df 90,227 4 11729,777 0,000
Based on trimmed mean 115,653 4 12010 0,000
z-Value: Deservingness Rationale Based on mean 79,964 4 12010 0,000
Based on median 74,925 4 12010 0,000
Based on median with adjusted df 74,925 4 11655,358 0,000
Based on trimmed mean 77,929 4 12010 0,000
z-Value: Economic Sociotropic Rationales Based on mean 26,381 4 12010 0,000
Based on median 22,486 4 12010 0,000
Based on median with adjusted df 22,486 4 11802,247 0,000
Based on trimmed mean 25,947 4 12010 0,000
z-Value: Cultural Sociotropic Rationales Based on mean 10,909 4 12010 0,000
Based on median 9,723 4 12010 0,000
Based on median with adjusted df 9,723 4 11951,407 0,000
Based on trimmed mean 10,694 4 12010 0,000
Tests the null hypothesis that the error variance of the dependent variable is equal across groups.
a. Design: Intercept + Welfare_regime_type
Despite the univariate test statistics significance, the SSCP Matrices indicated that the relationship between the outcome variables is the more important predictor for group differences, rather than the individual variables themselves (ibid.). Subsequent post hoc-tests23 shed more light onto the type of difference between regimes by examining differences between group means (ibid.). The Games-Howell test confirmed that there were significant differences between regime types with regard to the self-interest rationale. Regarding the deservingness rationale, there was no significant difference between the Anglo-Saxon and the Eastern, as well as between the Eastern and the Continental types, while between the others significant differences exists.
Meanwhile, with regard to the economic sociotropic rationale, significant differences existed merely with the Nordic regimes, while the other four do not show significant differences. Lastly, the cultural sociotropic rationale demonstrated significant differences between the Nordic, Southern and the other three regime types.
The REGWQ-test (Ryan-Einot-Gabriel-Welsch-Range) demonstrated homogenous subgroups and differences in a more graphic way (see Tables 7 through 10). It showed significant differences between all regime types regarding the self-interest rationale, with the Anglo-Saxon regime clearly taking up the most welfare chauvinist position, followed by the Southern, Continental, Eastern and, lastly, Nordic regime type; the latter three taking a more positive view. For the deservingness rationale, attitudes were likewise clearly distributed.
Here, the attitudes in the Nordic regime were the least welfare chauvinist, followed by the Southern regime, the Continental, Eastern and lastly Anglo-Saxon regime. The picture was far less clear-cut for the economic sociotropic rationale, where the Eastern, Anglo-Saxon and Southern regime were located at the end of welfare chauvinist attitudes, the Continental regime occupying the middle position and the Nordic regime type occupying the least welfare chauvinist position. Finally, for the cultural sociotropic rationale results showed the Anglo-Saxon and Continental regime at one end registering the most chauvinist attitudes, closely followed by the Eastern regime, while at the other end the Southern and, more prominently, the Nordic regimes are far less chauvinistic.
Table 9 - REGWQ-range economic sociotropic rationales
Means for groups in homogenous subgroups are displayed.
Basis: observed means.
The error term is mean of squares(Error) = ,972.
a. The critical Values for this dataare not monotonous. Substitutions were made to ensure monotony. Therefore, the Type I error is smaller.
b. Alpha = ,01
To gain a more detailed view of the differences between welfare regimes with regard to welfare chauvinist rationales, I followed up the MANOVA with a discriminant analysis (see Appendix 1.3). The latter revealed four discriminant functions, the first explaining 76,6% of the variance (canonical R² = 0,344)24, while the second, third and fourth explained 17,3% (R² = 0,172), 6% (R² = 0,102) and 0,1% (R² = 0,015) respectively.
Together, the discriminant functions differentiated the welfare regime types significantly, with Wilks’ Λ = 0,846, χ²(16) = 2003,05 and p = 0,000.25 Without the first function, the second function also significantly differentiated between welfare regime types; Λ = 0,960, χ²(9) = 488,108 and p = 0,000. The third function did so as well without the first and second, Λ = 0,989, χ²(4) = 128,097 and p = 0,000. The fourth function, meanwhile, did not significantly differentiate between the groups, Λ = 1,0, χ²(1) = 2,551 and p = 0,110. As shown by the correlations between outcomes and discriminant functions, the deservingness rationale loaded highly on the first function (r = 0,87) and reasonably on the second function (r = 0,47); the self-interest rationale loaded highly negatively on the first (r = - 0,68) and highly positively on the second function (r =
24 R² is the squared canonical correlation which serves as an effect size for the reported variance of the discriminant function (Field 2017: 561).
0,68). Meanwhile, both the economic sociotropic (r = - 0,48) and the cultural sociotropic rationale (r = - 0,41) loaded negatively on the first function and low on the second function. The economic sociotropic rationale, furthermore, loaded highly on the third function (r = 0,57) (see Tables 11 & 12). Lastly, as shown in the discriminant function plot, the first function discriminated the Nordic welfare regime from its neighbours, with the Anglo-Saxon regime building its counterpart. The Continental, Southern and Eastern regimes occupied a middle position, but clustered more closely to the Anglo-Saxon type. Meanwhile, the second function differentiated the Southern regime type from the Eastern and Continental regime type.
Table 11 - Standardised canonical discriminant function coefficients
Function
1 2 3 4
z-Value: Self-Interest Rationales -0,467 0,926 -0,386 -0,132
z-Value: Deservingness Rationales 0,753 0,620 0,007 0,459
z-Value: Economic Sociotropic Rationales -0,191 0,183 1,342 0,202
z-Value: Cultural Sociotropic Rationales 0,158 -0,419 -0,925 0,955
Table 12 - Discriminant function structure matrix
Function
1 2 3 4
z-Value: Self-Interest Rationales ,873* 0,474 0,044 0,106
z-Value: Deservingness Rationales -0,676 ,676* -0,218 0,197
z-Value: Economic Sociotropic Rationales -0,411 -0,153 -0,171 ,883*
z-Value: Cultural Sociotropic Rationales -0,482 0,087 0,565 ,664*
Pooled within-groups correlations between discriminating variables and standardised canonical discriminant functions Variables ordered by absolute size of correlation within function.
*Largest absolute correlation between each variable and any discriminant function
In sum, the MANOVA, despite remaining reservations, demonstrated that welfare regime types have a significant effect on distinct patterns of welfare chauvinist rationales. It most strongly distinguished the Nordic regime, which is significantly removed from the other regimes with regard to all outcome variables. Situated on the other side was the Anglo-Saxon regime type. The other regimes were located somewhere in the middle.
The discriminant analysis confirmed this pattern and could give additional insights into the relationship between the independent variables. It suggests that the two underlying dimensions best describing welfare regime differences were (1) the negative relation between the deservingness rationale and the other outcome variables, but especially the self-interest rationale; and (2) the differences in economic and cultural sociotropic rationales and the patterns of the self-interest and deservingness rationales. The results partly confirmed the starting hypothesis, that the pattern of welfare chauvinist rationales differs across welfare regime types (H2).
Similarly, it confirmed the expected pattern with regard to self-interest rationales, with the Anglo-Saxon as the most and the Nordic the least chauvinist regime (H2a). Regarding the deservingness rationale, the pattern likewise supported the prediction that this rationale is prominently expressed in Anglo-Saxon countries, less
so in the conservative regimes of Southern, Eastern and Continental Europe, while being far less notable in the Nordic countries (H2b). In contrast, the results for the sociotropic rationales did not follow the expected pattern. For the economic sociotropic rationale, the differences were small across regime types excepting the Nordic regimes, which expressed – contrary to my expectations – the least concerns about the economic impact of immigration for the wider community. The Eastern regime type featured more prominent economic concerns which similarly goes contrary to prior expectations. Accordingly, the sub-hypothesis could not be confirmed (H2c). Finally, the patterns for the cultural sociotropic rationale only partly confirmed prior assumptions (H2d): While the Nordic regime was the least concerned with immigration based on wider concerns about cultural cohesion, such concerns unexpectedly featured prominently for Anglo-Saxon regimes.
Correspondingly, while the Continental and Eastern regimes results conformed to expectations, the Southern regime was far less concerned with the cultural consequences of immigration than previously predicted.
After this report on the results of the first empirical analysis, I will now proceed with the second step of examining differences in the value base of welfare regimes. This may further illuminate possible causes explaining the unexpected deviations in the first part.